[Bugs] [Bug 1370092] New: VERY poor replica write perfomance

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1370092

            Bug ID: 1370092
           Summary: VERY poor replica write perfomance
           Product: GlusterFS
           Version: 3.8.3
         Component: replicate
          Severity: urgent
          Assignee: bugs at gluster.org
          Reporter: v.badalyan at open-bs.ru
                CC: bugs at gluster.org



Description of problem:
Replica 2 perfomance max is 500MB/s

Version-Release number of selected component (if applicable):
2 equal servers:
- E5-2690 x 2
- 256 GB 1600
- 40GBS infiniband link (Cx-2) p2p link. Connected mode.

How reproducible:
Create 2 RAM disk and create replica 2 volume. mount volume (TCP only, RDMA
only). Try to write. Only 400MB/s 

Steps to Reproduce:
on All servers:
- mkdir /mnt/ram
- mount  -t tmpfs -o size=1024M tmpfs /tmp/ram
- mkdir /mnt/ram/b1

gluster vol create ram replica 2 [transport...] s1:/mnt/ram/b1 s2:/mnt/ram/b1
gluster vol start ram

mkdir /tmp/glu
mkdir /tmp/ram
mount -t tmpfs -o size=1024M tmpfs /tmp/ram
mount -t glusterfs [-o transport=rdma] s1:/ram /tmp/glu

dd if=/dev/sda of=/tmp/ram/0.bin bs=128M count=5 #prepare data
dd if=/tmp/ram/0.bin of=/tmp/glu/0.bin bs=128M 

Result - 400-500MB/s
I was try do TCP tunning, use RDMA or TCP only, turn off all perfomance params
in vol. I can't get more 500MB/s. Ram disk read/write have 8GB/s
P.S. Many times try read from vol. Read is 500MB/s, 2GB/s, 2GB.

I have BIG and GREAT waits on write on my cluster! In production i use ovirt
hypercovered (3 servers, RAID0 8xSSD 1TB, Replica 3 arbiter 1)

Actual results:
Max write speed is 500MB/s

Expected results:
Write speed must be 6-8 GB/s

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.